Does the ordering temperature depend on how hot the simulation's scan goes?

EXPERIMENTS.md · lines 8444–8466

E138 — Is the rung-1 answer an artefact of the temperature range scanned?

E135's control failed: MoNbTaW read 589 K on a 200-2600 K ladder against E113's validated 724.9 +/- 23 K on a 200-2400 K one. The estimator fits its A/T^2 disordered tail on the hottest third of whatever range it is given, so changing the ceiling changes the fitting window and therefore the baseline. If that is the cause, every rung-1 number in this project is a function of a scan parameter, which no physical transition temperature can be.

Predicted:

1. MoNbTaW's reported transition moves monotonically with the ceiling - scanning to 2400, 2600 and 3000 K on identical seeds and sweeps gives three different answers, with the higher ceilings reading lower, because a hotter fitting window fits a smaller A and so subtracts less.
2. The spread across ceilings exceeds the 23 K seed scatter E113 measured, by a lot. E135 against E113 is already 136 K on one seed.
3. MoNbTaTiW stays above 1000 K at every ceiling, because the qualitative gap is about 800 K and no plausible baseline error closes it.

Falsified if the three ceilings agree within the seed scatter, in which case the 589 vs 724.9 discrepancy is something else - a change in rungs.transition, or seed variance larger than E113 measured - and the estimator is not grid-dependent after all. That would be the better outcome and it would move the problem, not remove it.